A Study of Image Detail Enhancement Using Modified Shock Filter and Progressive Image Denoising Algorithm

• Main Authors: Yen-Kai Huang, 黃彥凱
• Other Authors: 吳俊霖
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/4fb599

碩士 === 國立中興大學 === 資訊科學與工程學系 === 103 === Image detail enhancement is popularly a method of image processing. The method can boost fine scale features by increasing the contrast of the pattern on the object. This can make image clearer and informative. We decompose a single image into a smoothed mean layer and several detail layers. To avoid causing halo effects at edges, we use edge-preserving decompositions to capture detail layer. Some of the existing methods to achieve detail enhancement decompose the image into a smoothed mean layer and several detail layers based on Gaussian filter and bilateral filter suffer from halo effects, whereas methods based on weighted-least squares might cause noise-like structure enhanced. Based on local extrema filter can capture fine detail layer but might cause uneven enhancement. Base on L0 gradient minimization might not enhance some features if the gradients of the features are too large. In this study, we propose a new method to decompose an image using modified shock filter and progressive image denoising algorithm. We can set different parameters for multi-scale detail extraction. One algorithm of the method is modified shock filter combines principle of two filter, shock filter and Gaussian filter, to solve problems. We can capture fine detail to boost features. Another algorithm of the method is progressive image denoising algorithm. The detail layer by modified shock filter might exist noise-like structure when input image exists Gaussian noise. We use a hybrid method which deals with image in the spatial and frequency domain to reduce noise progressively on detail layer. In the experimental results, we compare our results with existing edge-preserving image decomposition algorithms and they demonstrate our propoed method achieves better performance.